[BlueX888]

Related Issue

Closes #4399

Description

A ChatAgent configured with response_terminators and a streaming model backend (model_config_dict={"stream": True}) never consulted its terminators: is_terminated() was never called, agent.terminated stayed False, and every yielded response carried terminated=False, so the agent never reported completion in stream mode while the identical non-streaming agent terminates on the same model output. Consumers that break their loop on terminated (e.g. camel/societies/workforce/role_playing_worker.py:154-162) can never end their chat via a terminator in this configuration.

Root cause: the only terminator call sites were the non-streaming ones — _step_impl (camel/agents/chat_agent.py:3132-3150), _astep_non_streaming_task (:3440-3458) and _step_get_info (:3959-3972). The streaming chunk processors _process_stream_chunks_with_accumulator / _aprocess_stream_chunks_with_accumulator never read self.response_terminators and hard-coded terminated=False in the final ChatAgentResponse (parent of this branch: :4868 and :5998), and _stream_response / _astream_response had no equivalent of the "Please continue." continuation loop that _step_impl runs when a terminator reports not-terminated. The constructor docstring (camel/agents/chat_agent.py:423-426) promises the agent "will keep prompting the model until a terminator signals completion" with no streaming carve-out, and the streaming code's own non-streaming fallback branch (via _convert_to_chatagent_response -> _step_get_info) already terminates correctly — the gap was only in the chunk-processing paths.

Changes (camel/agents/chat_agent.py only):

• Add _check_response_terminators, the same terminator pass _step_get_info performs (call is_terminated() on the completed message list, first firing terminator wins, update self.terminated).
• Call it at the record_message sites of the four streaming completion paths — _process_stream_chunks_with_accumulator, _aprocess_stream_chunks_with_accumulator, and the structured-output branches of _stream_response / _astream_response — and propagate the resulting status to the final streamed response instead of the hard-coded False. Partial responses keep terminated=False.
• In _stream_response / _astream_response, when a completed streamed response does not trigger a terminator, mirror _step_impl: stop with a warning at max_iteration, otherwise add the "Please continue." user message to memory and request the model again.

Out of scope: the streaming info dict keeps its existing schema (finish_reasons), so it still carries no termination_reasons entry when a terminator fires; that pre-existing schema difference is the secondary observation noted in the issue.

Testing

Reproduction first on unpatched master (24fde60a) with the four new regression tests added to test/agents/test_chat_agent.py (sync/async termination and sync/async continuation; fully offline via DummyModel + chunk stubs, no API key needed):

$ OPENAI_API_KEY=sk-dummy .venv/bin/pytest test/agents/test_chat_agent.py -k "stream_response_terminator" -q
FAILED test/agents/test_chat_agent.py::test_chat_agent_stream_response_terminators
FAILED test/agents/test_chat_agent.py::test_chat_agent_async_stream_response_terminators
FAILED test/agents/test_chat_agent.py::test_chat_agent_stream_response_terminator_continuation
FAILED test/agents/test_chat_agent.py::test_chat_agent_async_stream_response_terminator_continuation
4 failed, 38 deselected, 2 warnings in 1.13s

The core assertion failure is AssertionError: Expected 'mock' to have been called once. Called 0 times. — is_terminated() is never invoked in streaming mode. The issue's repro script also prints BUG PRESENT: terminator fired in non-streaming mode but was never consulted in streaming mode; agent never terminates. on the unpatched tree.

After the fix:

$ OPENAI_API_KEY=sk-dummy .venv/bin/pytest test/agents/test_chat_agent.py -k "stream_response_terminator" -q
4 passed, 38 deselected, 2 warnings in 0.71s

and the repro script now reports for the same scripted model output "Goodbye!":

non-streaming: response.terminated = True
streaming:    response.terminated = True (type: StreamingChatAgentResponse)
streaming:    agent.terminated   = True
streaming:    terminator fired   = True word_count_dict = [defaultdict(<class 'int'>, {'goodbye': 1})]

No existing test changes states:

$ OPENAI_API_KEY=sk-dummy .venv/bin/pytest test/agents/test_chat_agent.py -q
# unpatched master + new tests: 8 failed, 34 passed
# patched:                       4 failed, 38 passed
$ OPENAI_API_KEY=sk-dummy .venv/bin/pytest --fast-test-mode test/agents/test_chat_agent.py -q
14 passed, 28 skipped, 2 warnings in 1.09s
$ OPENAI_API_KEY=sk-dummy .venv/bin/pytest --fast-test-mode test/agents/ -q --ignore=test/agents/test_agent_e2e.py
3 failed, 77 passed, 49 skipped, 2 warnings in 6.51s

The 4 module failures (test_tool_calling_async, test_chat_agent_creation_methods, test_chat_agent_stream_with_structured_output, test_chat_agent_async_stream_with_structured_output) need a live OpenAI backend and fail identically on unpatched master. The 3 failures in the wider test/agents/ run (test_knowledge_agent.py::test_parse_graph_elements — ModuleNotFoundError for an optional dep, test_role_playing.py::test_role_playing_role_sequence, test_task_agent.py::test_task_planner_agent[model0]) also fail identically on unpatched master in this environment.

Lint and type check, run as CI runs them (.pre-commit-config.yaml):

$ pre-commit run --all-files
ruff...........Passed
ruff-format....Passed
Check mypy....Failed
Found 141 errors in 84 files (checked 852 source files)
Check License..Passed
codespell......Passed
(secrets / end-of-files / trailing-whitespace: Passed)

The mypy count is byte-identical to unpatched master in this environment (missing optional dependency stubs such as requests, playwright, nebula3, flask, pptx, sklearn); neither changed file reports an error.

What is the purpose of this pull request?

• Bug fix
• New Feature
• Documentation update
• Other

Checklist

• I have read and agree to the AI-Generated Code Policy (required)
• I have linked this PR to an issue (required)
• I have checked if any dependencies need to be added or updated in pyproject.toml and run uv lock — no dependency changes
• I have updated the tests accordingly (required for a bug fix or a new feature)
• I have updated the documentation if needed — not needed, existing behaviour only
• I have added examples if this is a new feature — n/a

AI assistance disclosure: this patch was drafted with AI assistance and reviewed by a human; the test output quoted above was produced by running the commands shown against this branch and unpatched master.

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

• Configuration used: Repository UI
• Review profile: CHILL
• Plan: Advanced
• Run ID: cf539b11-a5e7-4046-b7a5-e10bf67fee3c

---

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

---

_{Comment @coderabbitai help to get the list of available commands.}